We first consider the case in which tensoring with any object of \(C\) preserves the morphisms to be inverted.

Proposition 21.1.1 (Tensor-stable monoidal localization). Let \(C\) be a symmetric monoidal \(\infty \)-category, and let \(W\) be a class of morphisms in \(C\) such that, for every object \(X\in C\), the functor \(X\otimes -\) sends morphisms in \(W\) to morphisms in \(W\).

(1)

The localization \(C[W^{-1}]\) inherits a unique symmetric monoidal structure turning the localization functor \(\gamma \colon C \to C[W^{-1}]\) into a symmetric monoidal functor.

(2)

For any other symmetric monoidal \(\infty \)-category \(D\), restriction along \(\gamma \) induces a fully faithful functor \[ \Fun ^{\otimes }(C[W^{-1}],D) \hookrightarrow \Fun ^{\otimes }(C,D) \] whose essential image consists of those symmetric monoidal functors \(C \to D\) which invert the morphisms in \(W\).

Proof. We may replace \(W\) by the wide subcategory it generates. This leaves the localization unchanged, and tensoring with any object still preserves \(W\). Let \(\RelCat _\infty \subseteq \Ar (\Cat _\infty )\) be the full subcategory of relative \(\infty \)-categories, that is, inclusions of wide subcategories \(W\hookrightarrow C\). By the universal property of localization, the functor \((C,W)\mapsto C[W^{-1}]\) is left adjoint to \(D\mapsto (D,D^\simeq )\).

Both functors preserve finite products (Exercise 1.5.24), so this adjunction induces an adjunction on commutative monoids. Tensoring one variable at a time shows that the tensor product carries pairs of morphisms in \(W\) to morphisms in \(W\). Thus \((C,W)\) is a commutative monoid in \(\RelCat _\infty \). Applying the left adjoint gives a symmetric monoidal structure on \(C[W^{-1}]\), and the adjunction unit gives the symmetric monoidal functor \(\gamma \).

The adjunction identifies the anima of symmetric monoidal functors \(C[W^{-1}]\to D\) with the subanima of symmetric monoidal functors \(C\to D\) inverting \(W\). Apply this also to \(\Fun ([n],D)\) with its pointwise tensor product. A functor \(C\to \Fun ([n],D)\) inverts \(W\) precisely when its evaluation at every vertex does. We thus obtain the same identification on \([n]\)-diagrams for every \(n\), proving (2) at the level of functor categories (Definition 14.4.3). The universal property gives uniqueness in (1). □

Remark 21.1.2. In the situation of Proposition 14.7.2, we may consider the class \(W\) of \(L\)-equivalences in \(C\). The assumption on \(L\) guarantees that also the condition of Proposition 21.1.1 is satisfied. In particular, Proposition 21.1.1 provides an alternative way of equipping the subcategory \(D \subseteq C\) with a symmetric monoidal structure.

For chain complexes, we can compute the derived tensor product by first choosing projective resolutions (Definition 6.6.25). More generally, consider a tuple of cofibrant objects. After transport along a span, its entries are finite tensor products of cofibrant objects. To derive transport along a second span, we may have to replace these entries by cofibrant objects again. The following condition ensures that applying the second tensor product to these replacements gives a weak equivalence.

Definition 21.1.3 (Left derivable monoidal structure). Let \((C,W,I)\) be an \(\infty \)-category with weak equivalences and cofibrations, and suppose \(C\) is symmetric monoidal. We call an object tensor-cofibrant if it is a finite tensor product of cofibrant objects, allowing the empty tensor product. The monoidal structure is left derivable if for every finite family \((f_j\colon X_j\to Y_j)_{j\in J}\) of weak equivalences with each \(X_j\) cofibrant and each \(Y_j\) tensor-cofibrant, the map \[ \bigotimes _{j \in J} f_j \colon \bigotimes _{j \in J} X_j \to \bigotimes _{j \in J} Y_j \] is again a weak equivalence.

Remark 21.1.4. In the model category literature, there exists a notion of symmetric monoidal model category. The underlying category with weak equivalences and cofibrations of such a symmetric monoidal model category is always left derivable. However, the condition of being left derivable is much weaker: we require only that tensoring preserve the relevant weak equivalences, with no general compatibility between tensor products and cofibrations.

A convenient sufficient criterion is the following. Assume that the monoidal unit is cofibrant, that the tensor product of two cofibrant objects is again cofibrant, and that for every cofibrant object \(Z\) the functor \(-\otimes Z\) sends weak equivalences between cofibrant objects to weak equivalences. Then every tensor-cofibrant object is again cofibrant, and an induction on the cardinality of \(J\) shows that the monoidal structure is left derivable in the sense of the definition.

Construction 21.1.5. Let \((C,W,I)\) be an \(\infty \)-category with weak equivalences and cofibrations which comes equipped with a left derivable symmetric monoidal structure. For a finite set \(J\), let \(W^J\) denote the collection of morphisms in \(C^J \simeq C^{\otimes }_J\) given by \(J\)-tuples of weak equivalences, and let \(W^{\otimes }\) be the collection of morphisms in \(C^{\otimes }\) given by the union of the \(W^J\). By construction, the cocartesian fibration \(p_C\colon C^{\otimes } \to \Span (\Fin )\) inverts all morphisms in \(W^{\otimes }\), so the universal property of localizations produces a functor \[ p_{C[W^{-1}]}\colon C^{\otimes }[(W^{\otimes })^{-1}] \to \Span (\Fin ). \] The localization functor \(\gamma ^{\otimes }\colon C^{\otimes } \to C^{\otimes }[(W^{\otimes })^{-1}]\) then becomes a functor over \(\Span (\Fin )\).

Theorem 21.1.6 (cf. [Nikolaus and Scholze (2018), Theorem A.7]). Let \((C,W,I)\) be an \(\infty \)-category with weak equivalences and cofibrations, and assume that \(C\) comes equipped with a left derivable symmetric monoidal structure. Then the following statements hold:

(1)

The functor \(p_{C[W^{-1}]}\colon C^{\otimes }[(W^{\otimes })^{-1}] \to \Span (\Fin )\) defines a symmetric monoidal \(\infty \)-category.

(2)

The localization functor \(\gamma ^{\otimes }\colon C^{\otimes } \to C^{\otimes }[(W^{\otimes })^{-1}]\) encodes a lax symmetric monoidal functor.

(3)

The underlying functor of \(\gamma ^{\otimes }\) exhibits its target as the localization of \(C\) at \(W\), thus equipping \(C[W^{-1}]\) with the structure of a symmetric monoidal \(\infty \)-category.

(4)

For every symmetric monoidal \(\infty \)-category \(D\), the functor \(\gamma ^{\otimes }\) induces a fully faithful functor \[ \Fun ^{\otimes \textup {-lax}}(C[W^{-1}],D) \hookrightarrow \Fun ^{\otimes \textup {-lax}}(C,D) \] whose essential image consists of those lax symmetric monoidal functors \(C \to D\) that invert the morphisms in \(W\).

For chain complexes, the binary lax structure map \(\gamma X\otimes ^{\bL }\gamma Y\to \gamma (X\otimes Y)\) is the structure transformation of the derived tensor product. When \(X\) and \(Y\) are bounded below, it is represented by the map \(P\otimes Q\to X\otimes Y\) induced by a pair of projective resolutions.

Example 21.1.7 (Chain complexes). Let \(\Aa \) be a symmetric monoidal abelian category with enough projectives whose tensor product is additive in each variable. Assume its unit is projective and tensor products of projectives are projective. In the projective cofibration structure on \(\Ch ^-(\Aa )\), the cofibrant objects are the bounded below complexes of projectives (Proposition 6.6.15). They include the unit and are closed under tensor products, since only finite direct sums occur in each degree. Quasi-isomorphisms between them are chain homotopy equivalences, which tensoring preserves in each variable (Proposition 6.6.11, Definition 6.6.23). Hence the monoidal structure is left derivable.

By Theorem 21.1.6, the localization \(\Ch ^-(\Aa )\to \D ^-(\Aa )\) is lax symmetric monoidal. On bounded below complexes of projectives its structure maps are isomorphisms, since these complexes serve as their own resolutions. Thus localizing the symmetric monoidal subcategory \(\Ch ^-_{\proj }(\Aa )\) as in Proposition 21.1.1 gives the same symmetric monoidal structure on \(\D ^-(\Aa )\) by Proposition 6.6.16. Its tensor product is the one of Definition 6.6.25.

To prove Theorem 21.1.6, we describe what happens to cocartesian transport when we localize the fibers of a cocartesian fibration.

Recall from Definition 6.6.1 that a total left derived functor of \(F\colon C\to D\) is a right Kan extension of \(F\) along the localization \(\gamma \colon C\to C[W^{-1}]\). Such a derived functor \(\bL F\) is called absolute if for every functor \(H\colon D\to E\), the composite \(H\circ \bL F\) is again a total left derived functor of \(H\circ F\).

Example 21.1.8. Assume that \(F\colon C \to D\) inverts morphisms in \(W\), so that it induces a functor \(\bL F\colon C[W^{-1}] \to D\) by the universal property of localization. Then the natural isomorphism \(\bL F \circ \gamma \cong F\) exhibits \(\bL F\) as an absolute left derived functor of \(F\).

Example 21.1.9. More generally, let \((C,W,I)\) be an \(\infty \)-category with weak equivalences and cofibrations, in the sense of Definition 2.2.4, and consider a functor \(F\colon C \to D\) for which the restriction \(F\vert _{C_c}\colon C_c \to D\) to the cofibrant objects inverts weak equivalences. It follows from Proposition 6.6.3 that \(F\) admits an absolute left derived functor \(\bL F\colon C[W^{-1}] \to D\), uniquely characterized by the existence of an isomorphism \(\bL F\circ \gamma \circ i\simeq F\circ i\) on the cofibrant objects.

Definition 21.1.10. Let \(p\colon C \to S\) be a cocartesian fibration. Assume that for every object \(s \in S\) we are given a class \(W_s\) of morphisms in the fiber \(C_s := p^{-1}(s)\), and let \(\gamma _s\colon C_s \to C_s[W_s^{-1}]\) denote the localization functor. We say that \(p\) is left derivable with respect to \(\{W_s\}_{s \in S}\) if the following two conditions are satisfied:

(1)

For every morphism \(f\colon s \to t\) in \(S\), consider the cocartesian transport functor \(f_!\colon C_s \to C_t\). Then \(\gamma _t \circ f_!\colon C_s \to C_t[W_t^{-1}]\) admits an absolute left derived functor \[ \bL f_!\colon C_s[W_s^{-1}] \to C_t[W_t^{-1}]. \]

(2)

For each composable pair \(s \xrightarrow {f} t \xrightarrow {g} u\) in \(S\), the canonical map \[ \bL g_! \circ \bL f_! \to \bL (g \circ f)_! \] is a natural isomorphism.

To obtain the map in (2), compose the structure transformations to get \(\bL g_!\circ \bL f_!\circ \gamma _s\Rightarrow \bL g_!\circ \gamma _t\circ f_!\Rightarrow \gamma _u\circ (g\circ f)_!\). The universal property of the right Kan extension \(\bL (g\circ f)_!\) then gives the stated map.

Example 21.1.11. In the situation of Definition 21.1.10, assume that every cocartesian transport functor \(f_!\colon C_s \to C_t\) satisfies \(f_!(W_s) \subseteq W_t\). Then \(p\) is left derivable with respect to \(\{W_s\}_{s \in S}\). Indeed, the condition on \(f_!\) shows that \(\gamma _t \circ f_!\) uniquely factors through a functor \(\bL f_!\colon C_s[W_s^{-1}] \to C_t[W_t^{-1}]\), which by Example 21.1.8 is an absolute left derived functor. The second condition is then an immediate consequence of the universal property.

Example 21.1.12. If \(p\colon C \to S\) is left derivable with respect to \(\{W_s\}\), then for any functor \(f\colon T \to S\) the pullback functor \(C \times _S T \to T\) is left derivable with respect to \(\{W_{f(t)}\}_{t \in T}\).

Proposition 21.1.13 ([Nikolaus and Scholze (2018), Proposition A.14]). Let \(p\colon C\to S\) be a cocartesian fibration that is left derivable with respect to \(\{W_s\}_{s\in S}\). Write \(W\) for the union of these classes and \(q\colon C[W^{-1}]\to S\) for the functor induced by \(p\).

(1)

The functor \(q\) is a cocartesian fibration.

(2)

For each \(s\in S\), the localization functor induces an equivalence \(C_s[W_s^{-1}]\iso C[W^{-1}]_s\).

(3)

For \(f\colon s\to t\) in \(S\), write \(f'_!\) for transport along \(f\) in \(q\). Under the identifications in (2), the natural transformation \[ f'_!\circ \gamma _s\Rightarrow \gamma _t\circ f_! \] induced by \(\gamma \colon C\to C[W^{-1}]\) over \(S\) exhibits \(f'_!\) as the absolute right Kan extension of \(\gamma _t\circ f_!\) along \(\gamma _s\).

Proof sketch. When \(S=[0]\), the assertion follows from the definition of localization. When \(S=[1]\), the cocartesian fibration \(p\) is classified by a functor \(F\colon C_0\to C_1\). Unstraightening the absolute left derived functor of \(\gamma _1\circ F\) gives a cocartesian fibration over \([1]\) with the desired fibers. The description of its total category in Example 24.2.6, together with the absolute right Kan extension property, gives the universal property of the localization of \(C\).

Suppose now that \(S=\colim _i S_i\) and the result holds over each \(S_i\). Pullback preserves left derivability (Example 21.1.12). For each transition functor \(S_i\to S_j\), the map from the localized fibration over \(S_i\) to the pullback of the one over \(S_j\) is an equivalence on fibers by (2) and preserves cocartesian lifts by (3). The transition squares are therefore pullbacks. By descent for cocartesian fibrations, the colimit of their total categories is cocartesian over \(S\), and by the universal property of localization this colimit is \(C[W^{-1}]\). Part (3) extends to composites of arrows from the \(S_i\) by the composition condition in Definition 21.1.10. Since \([0]\) and \([1]\) generate \(\Cat _\infty \) under colimits (Proposition 25.1.14), this gives the result. See Nikolaus and Scholze (2018), Lemma A.16 and proof of Proposition A.14. □

Proposition 21.1.14. In the situation of Theorem 21.1.6, the cocartesian fibration \(p_C\colon C^{\otimes } \to \Span (\Fin )\) is left derivable with respect to the classes \(\{W^I\}_{I \in \Fin }\) on its fibers \(C_I^{\otimes } \simeq C^I\).

Proof. Fix a span \(\alpha \colon I \xleftarrow {f} K \xrightarrow {g} J\) in \(\Span (\Fin )\). Under the identification \(C_I^{\otimes } \simeq C^I\), the cocartesian transport functor along \(\alpha \) is given by \[ \alpha _!\colon C^I \xrightarrow {f^*} C^K \xrightarrow {g_{\otimes }} C^J, \qquad (x_i)_{i \in I} \mapsto \left (\bigotimes _{k \in g^{-1}(j)} x_{f(k)}\right )_{j \in J}. \] The product category \(C^I\) again carries weak equivalences and cofibrations componentwise, by Lemma 6.6.7. It therefore suffices, by Proposition 6.6.3, to show that \(\alpha _!\) preserves componentwise weak equivalences between componentwise cofibrant objects.

This is clear for the restriction functor \(f^*\). For the functor \(g_{\otimes }\), let \(\{u_k\colon x_k \to y_k\}_{k \in K}\) be a componentwise weak equivalence between componentwise cofibrant tuples. Then every object \(y_k\) is in particular tensor-cofibrant, and the \(j\)-th component of \(g_{\otimes }(u)\) is the map \[ \bigotimes _{k \in g^{-1}(j)} x_k \xrightarrow {\bigotimes _{k \in g^{-1}(j)} u_k} \bigotimes _{k \in g^{-1}(j)} y_k, \] which is a weak equivalence by the left derivability assumption. Thus \(\alpha _!\) admits an absolute left derived functor \[ \bL \alpha _!\colon C^I[(W^I)^{-1}] \to C^J[(W^J)^{-1}]. \]

It remains to verify compatibility with composition. Consider composable spans \[ \alpha \colon I \xleftarrow {f} K \xrightarrow {g} J \quad \qquadtext {and}\quad \beta \colon J \xleftarrow {h} L \xrightarrow {k} M. \] We show that the canonical map \[ \bL \beta _! \circ \bL \alpha _! \to \bL (\beta \circ \alpha )_! \] is an isomorphism. Let \(X=(X_i)_{i\in I}\) be componentwise cofibrant. Every component of \(\alpha _!(X)\) is a finite tensor product of cofibrant objects, and hence is tensor-cofibrant. Choose any componentwise weak equivalence \[ q\colon P\xrightarrow {\sim }\alpha _!(X) \] with \(P\) componentwise cofibrant. Such a map is obtained by factoring the morphism from the initial object to \(\alpha _!(X)\) in \(C^J\), and no functorial choice is required. By the defining property of total left derived functors, we have \[ (\bL \beta _!\circ \bL \alpha _!)(\gamma _I X) \simeq \bL \beta _!(\gamma _J\alpha _!(X)) \simeq \gamma _M\beta _!(P). \] The structure transformation for \(\bL \beta _!\) at \(\alpha _!(X)\) is represented under this identification by the map \[ \gamma _M\beta _!(q)\colon \gamma _M\beta _!(P)\longrightarrow \gamma _M\beta _!\alpha _!(X). \] Each component of \(\beta _!(q)\) is a tensor product of weak equivalences from cofibrant to tensor-cofibrant objects, so it is a weak equivalence by Definition 21.1.3. Thus the composite structure transformation of Definition 21.1.10 is invertible at \(X\), as is the structure transformation for \(\bL (\beta \circ \alpha )_!\) since \(X\) is componentwise cofibrant. It follows that \(\bL \beta _!\circ \bL \alpha _!\to \bL (\beta \circ \alpha )_!\) is invertible at \(\gamma _I X\). Every object of the localized fiber is represented by such an \(X\) by part (1) of Corollary 2.2.9. □

Proof of Theorem 21.1.6. Let us write \[ q\colon C^{\otimes }[(W^{\otimes })^{-1}] \to \Span (\Fin ) \] for the functor from Construction 21.1.5. By Proposition 21.1.14, the cocartesian fibration \(p_C\colon C^{\otimes } \to \Span (\Fin )\) is left derivable with respect to the classes \(\{W^I\}_{I \in \Fin }\) on its fibers. In particular, it follows from Proposition 21.1.13 that \(q\) is a cocartesian fibration. Moreover, for every finite set \(I\), the induced map on fibers \[ C_I^{\otimes } = C^I \to C^{\otimes }[(W^{\otimes })^{-1}]_I \] exhibits the target as the localization \(C^I[(W^I)^{-1}]\). Using Exercise 1.5.24, we obtain natural equivalences \[ C^{\otimes }[(W^{\otimes })^{-1}]_I \simeq C^I[(W^I)^{-1}] \simeq (C[W^{-1}])^I. \] For \(i\in I\), let \[ \rho _i\colon I\hookleftarrow \{i\}\xrightarrow {=}\{i\} \] be the corresponding inert span. Cocartesian transport along \(\rho _i\) in \(C^\otimes \) is the projection \(\pr _i\colon C^I\to C\), which preserves componentwise weak equivalences. Its total left derived functor is therefore the induced projection \[ C^I[(W^I)^{-1}]\longrightarrow C[W^{-1}]. \] Part (3) of Proposition 21.1.13 identifies this with cocartesian transport along \(\rho _i\) for \(q\). Consequently, under the equivalences above, the Segal map for the straightening of \(q\) is the equivalence \[ C^I[(W^I)^{-1}]\iso (C[W^{-1}])^I. \] Thus the cocartesian straightening of \(q\) preserves finite products, and Lemma 14.4.4 shows that \(q\) defines a symmetric monoidal \(\infty \)-category. This proves (1).

For (2), part (3) of Proposition 21.1.13 identifies the transformation induced by \(\gamma ^\otimes \) along \(\rho _i\) with the structure transformation of the derived projection. It is invertible since the projection preserves weak equivalences. Thus \(\gamma ^\otimes \) carries cocartesian lifts of \(\rho _i\) to cocartesian lifts. These lifts exhibit the product decompositions in both total categories by part (iii) of Proposition 14.1.12. Hence \(\gamma ^\otimes \) preserves finite products and defines a lax symmetric monoidal functor.

For (3), we evaluate part (2) of Proposition 21.1.13 on the one-point set \(\lra {1}\). This identifies the underlying \(\infty \)-category of the symmetric monoidal \(\infty \)-category from (1) with the localization \(C[W^{-1}]\), and under this equivalence the restriction of \(\gamma ^{\otimes }\) to the fiber over \(\lra {1}\) is the localization functor \(\gamma \colon C \to C[W^{-1}]\). Thus the symmetric monoidal structure from (1) may be regarded as a symmetric monoidal structure on \(C[W^{-1}]\) itself.

Finally, let \(D\) be a symmetric monoidal \(\infty \)-category. Since \(\gamma ^{\otimes }\colon C^{\otimes } \to C^{\otimes }[(W^{\otimes })^{-1}]\) is the localization of the total category at \(W^{\otimes }\), precomposition with \(\gamma ^{\otimes }\) induces a fully faithful functor \[ \Fun _{/\Span (\Fin )}(C^{\otimes }[(W^{\otimes })^{-1}], D^{\otimes }) \hookrightarrow \Fun _{/\Span (\Fin )}(C^{\otimes }, D^{\otimes }) \] whose essential image consists of those functors over \(\Span (\Fin )\) that invert the morphisms in \(W^{\otimes }\). In the sense of Definition 14.1.1, the lax symmetric monoidal functors are precisely the morphisms of \(\infty \)-operads, i.e. the functors over \(\Span (\Fin )\) that preserve finite products. Since \(\gamma ^{\otimes }\) preserves finite products, precomposition with \(\gamma ^{\otimes }\) certainly sends lax symmetric monoidal functors to lax symmetric monoidal functors. Conversely, let \(F\colon C^{\otimes }[(W^{\otimes })^{-1}] \to D^{\otimes }\) be a functor over \(\Span (\Fin )\) such that \(F \circ \gamma ^{\otimes }\) preserves finite products. By part (2) of Proposition 21.1.13, the functor \(\gamma ^{\otimes }\) is essentially surjective on each fiber, hence essentially surjective on the total category. Since it also preserves finite products, it follows that \(F\) preserves finite products as well. We conclude that the displayed fully faithful functor restricts to a fully faithful functor \[ \Fun ^{\otimes \textup {-lax}}(C[W^{-1}],D) \hookrightarrow \Fun ^{\otimes \textup {-lax}}(C,D) \] whose essential image consists of those lax symmetric monoidal functors that invert the morphisms in \(W\). Indeed, for a lax symmetric monoidal functor, inverting \(W\) on the fiber over \(\lra {1}\) is equivalent to inverting \(W^{\otimes }\) on every fiber, since the induced functor on the fiber over \(I\) is the product of the underlying functor on the fiber over \(\lra {1}\). This proves (4). □